ChatGPT was hardly a household name when Rep. Jacquelyn Baginski first pitched state regulations limiting artificial intelligence in elections in January 2024.
Seventeen months later, the AI-powered chatbot is part of the zeitgeist.
“Probably everyone has it downloaded on their phones right now,” Baginski, a Cranston Democrat, said in an interview Wednesday.
It’s only a matter of time, in Baginski’s view, before the rapid embrace of artificial intelligence filters through to local elections, potentially harming candidates and voters manipulated by deceptive images, audio and video created by generative AI. Which is why Baginski is again pushing for legislation that would restrict and regulate election deepfakes for the second year in a row.
Like last year, her proposal sailed through the Rhode Island House of Representatives Thursday, with a 64-1 vote, with one abstention.
Rep. Jennifer Stewart, a Pawtucket Democrat, cast the sole vote against the bill. Stewart did not immediately return inquiries for comment Thursday. Rep. Arthur Corvese, a North Providence Democrat, abstained. Corvese’s son-in-law works as policy director for Meta.
But across the rotunda, companion legislation from Sen. Lou DiPalma remains held for review by the Senate Committee on Judiciary, the same place where it languished and eventually died in the 2024 session.
DiPalma, a Middletown Democrat, insists he intends to shepherd the legislation “across the finish line” this year.
“This is an opportunity for us to get out in front of what’s going to happen,” DiPalma said in an interview Thursday. “It’s going to happen in Rhode Island. Let’s get out there and put some regulations in place before it does.”
Longstanding fears that AI could be used to manipulate elections and deceive voters was put to the test in 2024, with mixed results, according to experts. In January 2024, a robocall impersonating then-Democratic presidential candidate Joe Biden urged New Hampshire voters not to participate in the state’s presidential primary. The high-profile example of deceitful AI was later linked to a Democratic political consultant fined $6 million by the Federal Communications Commission.
But AI was used in positive ways too, helping candidates connect to voters more efficiently through self-deployed chatbots, emails and text campaigns. In Tokyo’s 2024 gubernatorial election, an obscure independent candidate placed fifth out of 56 contenders after using AI to respond to 8,600 submitted questions from voters, according to analysis by the Ash Center for Democratic Governance and Innovation out of the John F. Kennedy School of Government at Harvard University.
Secretary of State Gregg Amore acknowledges there are positive and negative outcomes for AI in elections. The legislation, which has Amore’s backing, doesn’t stop anyone from relying on artificial intelligence to advance campaigns or candidacies. Instead, it simply requires that the candidate, campaign or political action committee wielding AI disclose they are using manipulated images, audio or video in any ads that run within 90 days of an election.
“We’re not designing a felony here, we’re not trying to put someone away,” Amore said in an interview Thursday. “We’re just trying to make sure the public has accurate information 90 days before an election. And that they’re comfortable with the information they’re receiving – who it’s from and how it was created.”
Proactive, not reactive
It’s not unlike the 2012 state law that forced special interest groups to publicly report donations and spending to the state elections panel and include financial disclosures in any campaign ads, John Marion, executive director of Common Cause Rhode Island, wrote in a March 18 letter to the House Committee on State Government and Elections.
“We hadn’t yet seen large expenditures of dark money, but we knew we soon would,” Marion wrote, referring to the 2012 state law passed in reaction to the 2010 Supreme Court decision in the Citizens United case. “In 2014, that prediction came true and the state was ready to respond because of the General Assembly’s foresight.”
At least two dozen states have passed some form of regulations around AI advertising in elections, the bulk of them enacting laws in 2024 or later, according to an online tracker by D.C. think tank Public Citizen.
Rhode Island sought, but failed to join them in 2024, with the bill languishing in the Senate Committee on Judiciary.
Amore chalked up the stall to concerns raised by broadcasters and technology companies afraid of legal action from AI ads broadcast or posted through their platforms without proper disclosure. This year’s bill shifts the onus and risk of civil court charges from the content creator to the candidate, political action committee or campaign.
And it expands disclosure requirements to AI ads promoted via mobile applications and streaming services. Like last year’s iteration, satire and parody are exempt.
Despite these tweaks, the American Civil Liberties Union of Rhode Island again raised First Amendment concerns in a March 18 letter to lawmakers.
“While we recognize that free speech standards in the political arena are not limitless, this legislation, as worded, suggests that any image or recording that meets the definition of ‘synthetic media’ is deceptive or fraudulent and can therefore be regulated,” the ACLU wrote in its unsigned letter. “To allow the government to regulate or ban political speech that some might view as misleading undermines the breathing space that robust political speech requires, whether generated with the help of artificial intelligence or not.”
To allow the government to regulate or ban political speech that some might view as misleading undermines the breathing space that robust political speech requires, whether generated with the help of artificial intelligence or not.
Indeed, misinformation and deception in elections is a centuries-old malady, considerably worsened and more evident as technology has advanced, as documented in a 2022 report by the Bipartisan Policy Center. Ahead of the 2000 election, a humor magazine manipulated Google search results so that a search of “dumb motherf****r redirected to a George W. Bush merchandise center. The lighthearted “Googlebombing” devolved into more sinister tactics by 2016, when Russian interference into the U.S. presidential elections prompted a yearslong special federal investigation and report, resulting in 100 criminal charges to 34 individuals and 3 companies.
Given the rapid adoption of ChatGPT, Baginski and DiPalma don’t want to wait for the next iteration of technology to sow confusion in elections.
“It may not seem urgent yet because nothing has happened here yet,” DiPalma said. “Once something happens in a local election, watch how fast this passes.”
A committee vote on DiPalma’s bill has not been scheduled as of Thursday, according to Greg Pare, a Senate spokesperson.
Senate President Valarie Lawson remained noncommittal.
“I will be reviewing the committee testimony and speaking to the (Senate Committee) Chairman (Matthew) LaMountain regarding the particulars of this legislation,” Lawson said in a statement Thursday. “All elections should be conducted with the utmost fidelity and the highest standards of transparency and fairness.”
This story was originally published by the Rhode Island Current.
